When AI agents screen items for a costly human assessor, as in agentic search, e-discovery and AI reviewing of papers, only what they pass gets judged. We show what each kind of evidence can establish about their errors, even with unlimited data. Agreement shows that agents concur, not that they are right, however many there are. Acceptance by the assessor gives precision but only a lower bound on recall. Knowing which agents found which relevant items gives an upper bound on recall if agents' finds are not negatively correlated, and caps what one more agent can add if agents are interchangeable. Without assumptions about the agents, recall can be pinned down only by judging items they did not pass: a random sample of these estimates how many relevant items were missed, to any chosen accuracy, and the sample needed does not grow with the number of agents.
Accurate classifiers can fail when background cues change without altering the correct labels. Predictive performance alone does not quantify how much task information remains after accounting for such nuisance variation. Here we extend the Fisher-glass framework to connect nuisance-adjusted information with local classification difficulty and subpopulation auditing. We derive lower bounds on local minimax excess classification risk under explicit distinguishability and decision-separation conditions, alongside statistical audits accounting for estimation error and regularization. On Waterbirds, a feature-based diagnostic of task-background alignment selects an accurately classified group, with selection frozen before held-out evaluation. Across five training seeds, background replacements preserving the extracted bird foreground and labels yield 93.2% accuracy when background type is preserved and 40.6% when it changes, a controlled difference of 52.6 percentage points. Separately, search using task labels and a fixed nuisance direction identifies an independently evaluated population with 98.8% membership-weighted accuracy and low estimated projected predictive Fisher information. Training and acquisition experiments show that information gains and predictive gains can diverge. These findings support joint assessment of predictive performance, information under declared nuisance models and controlled sensitivity when diagnosing vulnerability and evaluating model improvements.
In ultra-high-resolution (UHR) remote sensing images, the objects that questions count in a region or relate by direction often span only a few pixels. Existing methods mainly decide where a vision-language model (VLM) should look and trust its answer in the chosen view. We show that this assumption fails for counting, and measure VLMs as one calibrates a sensor. All eight open VLMs tested fall increasingly short once a view holds more than a certain number of objects, about five for most on DOTA aerial images. This counting horizon narrows as views grow and objects appear smaller. An empirical power-law response model predicts the undercount a split recovers and the spurious counts it adds, and supports the rule of splitting only views beyond the horizon. In GeoScope, a text-only language model parses the question into a typed spatial program that fixes the region and its resolution. The VLM only grounds and counts within its operating range, and directions are computed explicitly from the grounded boxes. To our knowledge, it is the first UHR method to split views by a measured limit of the VLM. Without training, it improves Qwen3-VL-8B by 16.3 points on regional counting and 10.5 on spatial relations in XLRS-Bench, and has the highest reported accuracy, trained methods included, on regional counting and on the whole benchmark by question. On UHR-Micro it improves global and regional counting by 16.8 and 36.2 points, 15.6 of the former from splitting. On MME-RealWorld counting it gains 11.4 points, and we show that an option that is almost never correct yet often chosen distorts comparisons there. Project page: https://avalon-s.github.io/GeoScope/
Not all mistakes are alike. For example, in medical diagnosis, false positive and false negative mistakes have a different impact and should not be aggregated. By revisiting the classical theory of boosting through this lens, we introduce the notion of $Z$-learner, which returns weak hypotheses achieving joint false-positive/false-negative guarantees $(\mathrm{FP},\mathrm{FN})$ from some set $Z \subseteq [0,1]^2$. This formulation generalizes the standard notion of $\gamma$-weak learning--recovered by taking $Z=\{(z_1,z_2): z_1+z_2<1/2-\gamma\}$--and allows for more flexible notions of weak learnability that account for the impact of the distribution on the achievable learning guarantees. This raises a natural question: which sets $Z$ are boostable? We answer this by giving a complete geometric characterization that unifies, as special cases, both the classical boosting dichotomy and the recent multi-objective extension of Bressan et al. (COLT 2025). We also present an adaptive algorithm for boosting $Z$-learners without relying on the prior knowledge of $Z$, generalizing AdaBoost's ability to handle unknown margin parameters. Finally, we demonstrate that $Z$-learners can achieve confidence amplification. However, the analysis is significantly more involved than in classical boosting and relies on the centerpoint theorem from discrete and computational geometry.
Woody breast (WB) myopathy is a spatially heterogeneous fibro-degenerative condition of the broiler Pectoralis major muscle. Existing mechanical reference methods rely on single-point or sequential compression measurements, limiting characterization of spatial hardness variability. This study presents TACTUS (Tactile Adaptive Conformal 3D Uniform Sensing), an automated tactile array that generates thickness-normalized, three-dimensional hardness maps at 20 anatomically distributed locations within approximately 30 s. TACTUS integrates 20 independently motorized tactuators with force-sensing resistors and encoder-driven linear actuators, alongside a LiDAR depth camera for surface profiling and adaptive compression. The system was evaluated using 100 broiler breast fillets across four WB severity classes (Normal, Mild, Moderate, Severe). TACTUS identified significant force differences among severity classes (ANOVA, p < 0.001), with the strongest differentiation in the cranial region and a significant caudal effect not detected by texture-analyzer measurements. Fillet-level TACTUS force showed moderate but significant correlation with texture-analyzer measurements (Pearson r = 0.441, p < 0.001). Semivariogram, local change entropy, and heterogeneity analyses revealed increasing and anatomically structured hardness variability with WB severity. These findings demonstrate TACTUS as an automated, operator-independent platform for spatially resolved mechanical characterization and ground-truth generation to support hyperspectral and near-infrared spectroscopy-based WB classification systems.
We study the problem of binary classification in the marginal-nonuniform learning framework in the agnostic setting. In this model, the constants in the learning rate may depend on the marginal distribution of the unlabeled data, but the guarantee must hold uniformly over all conditional distributions of the labels. We also study fine-grained rates. We consider two criteria. (1) The standard agnostic setting, where the learner competes with the best concept in the class. In this setting, an immediate consequence of standard agnostic PAC learning results gives us the optimal rate of $\tfrac{1}{\sqrt{n}}$ for every class $\mathcal{H}$ with finite $\mathrm{VC}(\mathcal{H})$ and shows that arbitrarily slow rates are necessary otherwise. This also gives us the optimal fine-grained rate of $\sqrt{\tfrac{\mathrm{VC}(\mathcal{H})}{n}}$. Therefore, the dependence on the marginal distribution does not help, which is in contrast to the realizable setting. (2) In the relaxed agnostic setting, where the learner competes with $(1+\lambda)$ times the error of the best concept for a fixed $\lambda \in (0,1]$, the picture from the realizable setting reappears. We obtain a trichotomy of the achievable optimal rates: exponential rates of the form $e^{-n}$ arise precisely when the class is finite; linear rates of the form $\tfrac{1}{n}$ arise when the class is infinite but $\mathrm{VC}(\mathcal{H})$ is finite; and arbitrarily slow rates occur when $\mathrm{VC}(\mathcal{H})=\infty$. We also obtain the optimal fine-grained rate of $\tfrac{\mathrm{uVCE}(\mathcal{H})}{\lambda n}$, where $\mathrm{uVCE}(\mathcal{H})$ is a recently introduced combinatorial complexity dimension called the uncentered $\mathrm{VC}$-eluder dimension.
Ottoman Divan poetry is composed in aruz, a quantitative meter based on the weights of syllables. The meter is sensitive to specific characters recorded in the original writing in the Arabic script. This work is concerned with automatic meter finding and scansion alignment for Ottoman Divan poetry, especially on modern Turkish spelling without the special diacritics that scholarly transliterations of Ottoman preserve. We define two tasks on this input, first identifying the meter used in the poem, and second aligning the meter's weights with the syllabification of the poem, identifying the flaws and freedoms the poet has taken along the way. We build a corpus of 52,000 lines split by poet, with scansion annotations for 500 test lines, and a robustness benchmark generated at the deviation rates measured on real web and OCR copies. Our model is a hidden Markov model over the positions of each meter, decoded over a lattice of every candidate path, and trained by EM from the printed meter, without scansion labels. On a held-out set it scans 77.6% of the test lines fully correctly, against 26.8% for the available rule-based scanner, 55.0% for a neural tagger trained with the same supervision and 63.6% for Claude Opus 5.5 few-shot. The code is open-sourced and the tool is released on the web.
Causal inference under treatment interference requires assumptions to reduce the exponentially large number of potential outcomes. We develop hierarchical finite-population models for potential outcomes by separating these assumptions into three layers: interference neighborhoods, exposure mappings, and outcome models. Given an interference neighborhood and exposure mapping, we show that potential outcomes admit a nonparametric decomposition into direct, spillover, and treatment--spillover interaction components. This decomposition provides a common coordinate system for nonparametric, semiparametric, and parametric models of finite-population potential outcomes. For a given potential-outcome model, we define two classes of estimands. Average causal effects (ACEs), such as direct and total treatment effects, contrast potential outcomes under fixed treatment--exposure conditions. Expected causal effects (ECEs), such as expected direct and indirect effects, contrast expectations of potential outcomes under a specified allocation policy. For both classes, we bound the variance of the canonical Horvitz--Thompson estimator by three terms: a diagonal weight term, a term from overlapping interference neighborhoods, and a residual design-dependence term. Our upper and lower bounds show that under graph-unaware designs, nonparametric design-unbiased estimation of ACEs can require sample sizes growing exponentially in the interference degree, whereas polynomial growth can suffice for design-aligned ECEs. Under neighborhood interference, we construct graph-aware independent-set designs with variance bounds polynomial in the maximum degree for direct and total treatment effects on every graph, whereas graph cluster randomization can incur worst-case variance exponential in the degree.
Stellarators are fusion devices that use externally generated, three-dimensional magnetic fields to confine hot plasma and offer a promising route to commercial fusion. Their electromagnetic coils pose a challenging design problem: they must reproduce the target magnetic field while satisfying engineering constraints and remaining practical and cost-effective to manufacture. Finding suitable geometries is a highly nonlinear inverse problem that generally has neither a closed-form nor a unique solution. This problem lies at the heart of stellarator design. We introduce CoilStellaration, a dataset and benchmark for learning the mapping between coil geometries from target plasma magnetic fields and engineering requirements. The dataset contains 177,000 optimized filamentary coilsets targeting plasma equilibria from ConStellaration, with sampled engineering requirements and achieved geometric and magnetic-field metrics. It spans multiple coil counts and field-period numbers, supporting studies of both specialized and shared models across configurations. The benchmark asks models to predict coils conditioned on a target equilibrium and requirements on coil-coil and coil-plasma spacing, curvature, and magnetic-field error. We show that magnetic-field accuracy remains the main challenge, with learned models rarely satisfying all requirements simultaneously. However, using learned predictions to initialize conventional coil optimization can accelerate early progress. We release baseline models and evaluation code to support reproducible research on both direct coil generation and learned optimization initializations. By lowering the barrier for machine learning and optimization researchers to engage with stellarator coil design, we aim to accelerate progress toward practical, cost-effective stellarator power plants.
Prompt-based image generation makes it easy to produce visual assets, but offers little support for engaging with the creative process. Prompting collapses interdependent creative decisions into a monolithic instruction and returns a finished image, leaving control limited to revising the prompt or correcting the model's output. We present reComposer, a process-oriented image-generation interface built around editable planning, which decomposes users' evolving intent into meaningful creative decisions - such as subject, composition, or palette - and collaborates with users to explore underspecified goals and alternatives. reComposer supports conversational and direct manipulation of the plan, moving between concept-level intent and visual-level control. We evaluated reComposer with 8 design and non-design professionals. Compared to prompting, participants felt greater control and ownership while developing visual assets, exploring alternatives, and iterating on details. Our findings suggest how shifting generative AI interaction from specifying outputs to shaping process can support creative thinking alongside asset production.
Medical images encode intrinsic tissue properties that indicate disease progression, yet these properties remain entangled with the system-dependent physics of image formation. Differentiable inverse rendering asks whether they are recoverable by inverting the imaging system's forward physics rather than through post-hoc correction. We address this in optical coherence tomography (OCT), focusing on the optical attenuation coefficient, a per-pixel marker of pathology and cellular composition. Conventional recovery relies on closed-form inversions limited to simplified light-tissue models, or learned methods coupled to specific imaging configurations. We instead recover per-pixel attenuation from a single OCT image by gradient descent through a differentiable physics-based forward model. We verified the method against analytical solutions, validated it on five in vitro phantoms, and applied it without modification to 29{,}269 clinical intravascular OCT images from 128 patients. The inversion needs no training data or segmentations and is portable across platforms and retrospective archives. Replacing the forward model or recovering another optical parameter leaves the optimization procedure unchanged. We demonstrate per-pixel recovery is tractable because gradient information scales to this high-dimensional problem, where gradient-free methods become intractable. Our work supports quantitative tissue characterization from OCT images and provides a framework for extension across imaging modalities and organ systems.
Gastroenteropancreatic neuroendocrine tumors (GEP-NET) lack prognostic biomarkers, and computed tomography (CT)-based body composition radiomics at the third lumbar vertebra (L3) is a promising candidate. Small cohorts make high-dimensional radiomic models prone to overfitting and single-split evaluations variable and optimistic. This methodological study proposes a four-step validation protocol for radiomic survival analysis in small cohorts: (1) false discovery rate (FDR)-corrected feature selection, (2) a dual-path comparison of prior feature selection versus embedded regularization, (3) repeated nested cross-validation (CV) and (4) a negative control with Gaussian noise features. In 78 GEP-NET patients, 535 radiomic features from five L3 compartments were used to evaluate five survival models with Harrell's concordance index (C-index), time-dependent area under the curve (AUC) and integrated Brier score (IBS). Step 1 selected at least one feature in 18% of outer folds under noise (nominal level 20%) versus 84% with radiomic features. The dual-path comparison showed no consistent difference between strategies. Repeated nested CV reduced variability across seeds (41% median reduction in C-index standard deviation), but not enough to distinguish models. The nested CV C-index exceeded chance by 0.07-0.15 with radiomic features but not clearly under noise (0.40-0.42 with prior feature selection, 0.47-0.55 with embedded regularization). Frequently selected features tentatively linked greater subcutaneous fat, skeletal muscle compactness and texture to better prognosis. These findings suggest a modest prognostic signal in L3 radiomics, with considerable uncertainty that the protocol quantifies. The protocol may transfer to other small-cohort radiomic survival studies. Code: https://github.com/marielatalaveram/radiomics-survival-validation-protocol.
Recent geospatial foundation models have introduced a new paradigm for large-scale Earth observation by providing pretrained embedding representations that can be directly exploited for downstream tasks. In this work, we investigate the use of Google AlphaEarth Foundations embeddings for national-scale crop type mapping in Morocco. The proposed workflow operates directly on publicly available 64-dimensional pretrained embeddings using lightweight downstream classifiers, eliminating the need to process raw satellite imagery during model training and inference. The approach targets three crop categories, namely winter cereals, tree crops, and other crops. For nationwide inference, the Fields of the World global field boundary map is used to constrain predictions to agricultural parcels, enabling efficient large-scale mapping. Experimental evaluation on an independent test set over ten repeated runs achieved an average classification accuracy of (87.17 +/- 0.42%). Nationwide inference over Morocco was completed in approximately five hours, with the dominant computational cost arising from cloud-based retrieval of the embeddings rather than model execution. These results demonstrate the potential of pretrained embedding representations for computationally efficient national-scale crop type mapping.
A world model predicts how a scene changes after an action, and a planner uses it to search for actions that reach a goal. If the camera moves or the controller changes, the same command moves the robot differently: when its commands are turned by 90 degrees, a frozen planner with the LeWM world model that succeeds in 77.5% of episodes succeeds in only 7.8%. Current fixes retrain the model, need training data, or forget a set-up when it returns. We propose Library of Hands (LoH), a training-free method that keeps the world model frozen and corrects only the map from commands to applied actions, which we call the hand. LoH reads the hand through the frozen model, with no extra sensor: it finds the action that best explains the change between two camera images (a reading). It stores every hand it identifies, so a returning set-up is recognised rather than re-learnt. In closed-loop tests, LoH more than doubles success on three simulated tasks, from 28.6% to 62.3%, and loses only 0.6 percentage points when nothing changes. It also works, unchanged, with two other world models and on two of three models pretrained on real robot data, and its gain grows with the quality of its readings.
Vision-language-action (VLA) models are trained as general-purpose foundation models for robot learning, and require fine-tuning for each downstream task and robot. This raises practical questions such as how much data and training updates are needed, how checkpoints should be selected, and whether simulation can be leveraged to alleviate the cost of real data collection. We study these questions on a task that requires both precise manipulation and spatial understanding of the scene. In this task, an SO-101 arm moves a chess king between two squares that are given only by their names, such as e4 and a5. We fine-tune $\pi_{0.5}$ with various dataset sizes and update budgets, using teleoperated demonstrations as well as simulated demonstrations from a randomized scripted expert that imitates the variability of human demonstrations. In simulation, the success rate scales with both data and updates. Importantly, we find that a short learning-rate cool-down adds up to 31 points and is necessary for comparing checkpoints, and that validation loss and action error are unreliable for model selection. On the real arm, we fine-tune on up to 256 teleoperated demonstrations and evaluate ten policies in 25 trials each. Partial results indicate that pretraining on the simulated demonstrations helps substantially, most clearly when real data is scarce. We also present a negative result, in which training on the policy's own rollouts with filtered behavior cloning, RECAP-style advantage conditioning, corrections from an automatic coach, or residual reinforcement learning did not yield a measurable improvement over plain behavior cloning in our setting. We release the simulator, the expert and the real demonstrations.
Interpretability tools are usually built for one model at a time: a refusal direction, a truthfulness probe, or a steering vector is found with labeled data inside the model it explains. We ask whether such tools can be carried to other models through a map between residual streams fit only on unlabeled prompts. We transfer a refusal direction, tested causally by addition and ablation, and two read-out tools, a correctness direction and a sentiment probe, across 110 pairs of models from five families and 1B to 72B parameters and between base and instruction-tuned models, with controls for specificity and five seeds. How much transfers depends on what the map was fit on. A map fit on ordinary questions recovers 54\% of the target's own refusal steering and 16\% to 77\% of its correctness gain; adding unlabeled prompts from the region where the tool operates raises these to 102\% and 88\% to 98\%, while control directions recover at most 1\%. We prove that a linear map behaves this way whenever two models correspond linearly on the data it covers, and a label-free coverage score derived from this result tracks recovery across more than 16,000 runs and explains earlier reports that maps between models fail across domains. Carried read-out tools outperform anything the target's unlabeled representations yield. Interpretability results, once found, can be reused across models at the cost of a few unlabeled prompts.
Scribble-supervised segmentation is typically evaluated with a fixed annotation set, leaving robustness to distinct semantic and geometric imperfections poorly understood. Motivated by an audit of 12,031 manual PASCAL Visual Object Classes (VOC) 2012 scribbles and prior work, we introduce six controlled perturbation operators. Class corruption, spatial offset, and directional perturbation directly operationalize audit signals. Stroke-length reduction is motivated by prior scribble studies, while random dropout and fragmentation serve as diagnostic controls. Each operator reports requested and achieved changes. We evaluate manual and automatic scribbles across two image domains, two principal backbone settings, and multiple training pipelines, with an additional MiT-B1 PFA anchor evaluation. In the five-seed Tree Energy Loss (TEL)-ResNet101 VOC study, mean intersection over union (mIoU) drops from 70.08 to 1.46 under class corruption, compared with 35.57 for spatial offset, 52.20 for directional perturbation, and 63.24 for stroke-length reduction at their strongest settings. At the evaluated endpoints, semantic-anchor corruption produces the largest degradation, while geometric and support responses depend on the pipeline and dataset. Our benchmark provides a reproducible protocol for diagnosing annotation robustness beyond clean-set accuracy. The operator code is available at https://github.com/Wmhiram99/scribble-operator-benchmark.
This paper introduces a geometric approach to the representation and learning of time-variant parametrizations of deep assignment flows, for binary data labelings on graphs. Each edge carries an optimally controlled parameter flow in a 2-dimensional totally geodesic submanifold, based on objectives derived from a non-local linear scale space representation of the input data. The aggregation of triplets of such flows in 5-dimensional totally geodesic submanifolds is learned from data. The fusion of these flows determines the trajectory in the high-dimensional symmetric parameter space and by geometric integration the layers of binary deep assignment flows. Numerical results demonstrate the expressivity and the limitation of our highly parallel geometric local synthesis approach to high-dimensional deep assignment flow parametrizations.
Trade promotions consume a large share of consumer-goods revenue, yet most planners still set discounts with a single, often unreliable, price elasticity per item. We present an end-to-end Trade Promotion Optimisation (TPO) system built as one chain: analogue SKUs, point elasticity, elasticity curve, price recommendation, promotional lift and net lift. Elasticity is measured at post-promo time from directional price-dip events against a rolling regular price, and at promo time by a causal estimator that residualises trend, seasonality and promotion, handles stock-out censoring with a Tobit likelihood and updates itself from the first clean promotion week. Items with weak evidence borrow the point elasticity and the curvature from analogue SKUs chosen by a cross-validated ridge model over category, brand, sub-category, price line and description. The optimiser separates self-price lift from substitution and post-promo dips and adjusts for inventory and trade funding. In a four-week promotion experiment on a 39-SKU consumer brand (mean elasticity -3.06), the plan lifts units by 20.6% and revenue by 6.9% at flat margin. Analogue pooling halves elasticity error relative to hierarchy averaging.
Routing a melanoma case to routine follow-up is a serious action error that diagnostic coverage alone does not rule out. We present Harm-Constrained Risk-Optimal Conformal Prediction (HC-ROCP), a training-free framework that uses frozen classifier probabilities and adds a harm penalty when choosing prediction sets and actions. We test the resulting policies on disjoint calibration splits with multiplicity adjustment. HC-ROCP certifies supported claims, reports descriptive fallback policies, and refuses unsupported subgroup claims. We evaluate these outputs retrospectively on COVID-19 radiographs, ISIC 2019 dermoscopy, and EyePACS-to-APTOS retinopathy transfer. Across ten calibration seeds, HC-ROCP reduces COVID-19 severe-action errors from the exact risk-optimal baseline's 4.68\% to 0.54\%. At a 5\% budget, global claims are certified for COVID-19 and ISIC, but the rare squamous-cell-carcinoma claim is refused. At 1\%, all settings receive descriptive reports. For each budget, we also report the minimum severe-case count needed to pass the global harm gate with zero errors, providing a starting point for further data collection.